Research On Traffic Sign Recognition Method Based On Deep Learning

Recent years have seen a significant advancement in deep learning technology as well as a range of target identification algorithms,which has led to an increase in the relevance of technological advancements in real-world applications.In the use of intelligent transportation systems,technology is crucial.Driving assistance and timely road condition information are made possible by traffic sign detection and identification technology.How to identify the traffic signs quickly and accurately in the process of driving is always the key and difficult problem in the process of research.Convolutional neural network-based methods are shown to handle information more quickly and with higher accuracy than conventional detection algorithms.Have better research value.In an effort to address the issues of false and missing detection while identifying small objects in the existing YOLOv5 network,this paper studies and improves the following aspects based on the YOLOv5 network:In the beginning,the benefits and drawbacks of the four YOLOv5 networks are examined alongside real-world application scenarios.The YOLOv5 s network,which has the best real-time performance in YOLOv5,is selected as the basic network for this study,and the coordinate attention mechanism is embedded into the feature extraction network to improve the network’s accurate positioning and attention to traffic sign targets.Ensure the network to extract effective feature information of small target.At the same time,a cross-layer connection is added on the basis of the feature aggregation path to reduce the loss of feature information in the multi-layer transmission,and more features are fused together.Studies reveal that the enhanced YOLOv5 s network model,when applied to the GTSDB traffic sign data set,can accurately recognize and classify 43 different types of traffic signs with only a slight volume increase.but there are still problems of missing detection of small target signs and identification errors between similar signs,and there is still room for improvement.Secondly,replacing the model with the Ghost module after the first layer of all subsequent convolution operations can help the ResNet50 network,a single fine classification network,improve.so that part of the original network characteristics of nonlinear transformation into linear transformation,Improve classification speed by reducing the amount of parameters and model calculations.Finally,For traffic sign detection and classification,the enhanced YOLOV5 s network is connected to the compact ResNet50 network.First,The enhanced YOLOv5 network was utilized to recognize and categorize traffic signs into four groups,and the results were input into the cascaded lightweight ResNet50 network for 43 categories of fine classification.Compared with the single improved YOLOv5 s network,the cascaded network has increased the size of the model,but its detection accuracy on the GTSDB data set has been improved,and it can complete the task of rapid recognition and classification of traffic signs.